Modern commodity hardware is based on a single platform. Generally, it is manufactured by Intel or AMD, and is typically x86 architecture. The instruction sets have changed little except for additional instruction sets for Graphics processing and virtualization.

The modular hardware concept models software development practices. Good software is modular. It allows you to upgrade it, change components and add or remove functionality without impacting or limiting other features or capabilities of the greater system.

Our hardware platforms should be no different. The manufacturers of servers can make the hardware with pluggable FPGA and ASIC modules that support common, specific functionality such as running threads of a Web server or application server, or performing relational algebraic algorithms for data stores or key value stores. There is no longer a need for one-platform-fits all.

Many proprietary network devices benefit from this design, although they are not modular. A network-based load balancer usually employs the work of an ASIC or FPGA to perform functions that are frequently repeated, such as managing connections or performing quality of service calculations. It uses an FPGA or ASIC because it does one thing exceptionally well, and extremely fast. I propose to the technical community at large, and especially hardware manufacturers that we begin discussing such a design to achieve orders of magnitude in performance over our existing design.

Most servers these days are accessible directly or indirectly over the Internet. The Internet is how we communicate. Most of our Internet services are exposed to the end user in the form of a Web site (application) or consumable via a Web service, or API. We can keep our monolithic server designs, but for the most commonly used functions, we can begin to employ modular hardware architectures as it applies to servers to make a quick leap in performance of our most common systems.

The other considerations are in regard to the environmental and recycling cost of producing chasses for commodity hardware. There is a cost in recycling chasses, or mining ore to produce new chasses. By modularizing the main components of the system, including the bus architecture, one could reduce the environmental impact associated with producing and reproducing systems on the global scale of refresh of commodity hardware. The other savings in dollars as well as environmental cost is the transport cost associated with moving large machines over air and land as opposed to small modules.